Perplexity's free AI sorts and labels documents faster than a chatbot
Perplexity has released Decider, a downloadable AI model that answers multiple-choice questions about text and images in one pass. It suits high-volume jobs like routing, tagging and screening.

The short version
- Perplexity has released Decider, a free AI model built only to sort and label things.
- It answers multiple-choice questions about a document or image in one quick pass, instead of writing a reply.
- On Perplexity's own 63 sorting tasks, it was right 91.6% of the time, beating the tools Perplexity already uses.
What Perplexity released
Perplexity released Decider on 12 November 2026. A smaller, quicker version called Decider Flash came with it. Both are free to download from Hugging Face under the Apache license. They are also available as a paid service.
Decider does one job. You give it something to read. That could be an email or a screenshot. Then you ask it questions with fixed answers. Is this a complaint or a question? Is this page useful or spam? Does this document contain personal data? For each question, it returns how likely each answer is.
A new feature called locate also pulls exact text out of a document. Ask for "the city where the job is based" and it returns the matching words and where they sit.
Why this matters
Many companies use large chat AI tools for jobs like these today. That works. But at high volume it is slow and expensive. A chat model writes its answer word by word, even when the answer is just "yes".
Decider skips the writing. It reads the input once and answers several questions about it at the same time. Perplexity says that makes it considerably faster and cheaper than using a chat model to judge. It also gave steadier answers. When Perplexity shuffled or renamed the answer options, Decider changed its answer on just 0.27% of questions.
Perplexity shared results from its own use:
- Sorting search requests. Across 63 of its own tasks, Decider was right 91.6% of the time, measured against answers checked by another AI model. That beats the tools Perplexity was already using by 6.54 points.
- Checking its AI helpers. In one day, Decider read through 7,000 sessions of Perplexity's AI helpers at work. It found gaps that led to 14 proposed rule fixes, 6 of which shipped within two days. Replayed on a week of records, the fixes cut the busiest alert rule's volume by 52%.
- Cleaning its search index. Decider scores each crawled web page for quality. In the same pass, it also flags personal information.
Perplexity also tested Decider in eight languages. Japanese and Korean were among them. Indonesian was not.
What it means for companies in Indonesia
Most companies have piles of small decisions that a person makes by reading something. Is this ticket urgent? Which team should handle it? Does this invoice match the order? Is this ad copy on brand?
A tool like Decider can make these calls quickly and in bulk. Each answer comes with a confidence score, so you decide how sure it must be before acting. Because it is free to download, you can also run it on your own servers, which keeps sensitive data inside the company.
You still need to set it up with care. Perplexity itself recommends tuning the threshold on a small set of your own labeled examples. A false alarm costs different amounts in different jobs.
Language is the main gap. Test it on Bahasa Indonesia material before you rely on it.
What to try
- List the yes-or-no and pick-one decisions your teams make by reading documents every day.
- Pick one high-volume task such as routing customer emails. Label a few hundred real examples of it.
- Test both versions against those labels, using Indonesian text as well as English.
- Set the threshold based on what a wrong answer costs, and keep a person checking the uncertain cases.
